Robust Disambiguation of Named Entities in Text

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 782–792, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Robust Disambiguation of Named Entities in Text Johannes Hoffart1, Mohamed Amir Yosef1, Ilaria Bordino2, Hagen F¨urstenau3, Manfred Pinkal3, Marc Spaniol1, Bilyana Taneva1, Stefan Thater3, Gerhard Weikum1 1 Max Planck Institute for Informatics, Saarbr¨ucken, Germany 2 Yahoo! Research Lab, Barcelona, Spain 3 Saarland University, Saarbr¨ucken, Germany {jhoffart,mamir,mspaniol,btaneva,weikum}@mpi-inf.mpg.de bordino@yahoo-inc.com {hagenf,pinkal,stth}@coli.uni-sb.de Abstract Disambiguating named entities in natural- language text maps mentions of ambiguous names onto canonical entities like people or places, registered in a knowledge base such as DBpedia or YAGO. This paper presents a ro- bust method for collective disambiguation, by harnessing context from knowledge bases and using a new form of coherence graph. It unifies prior approaches into a comprehensive frame- work that combines three measures: the prior probability of an entity being mentioned, the similarity between the contexts of a mention and a candidate entity, as well as the coherence among candidate entities for all mentions to- gether. The method builds a weighted graph of mentions and candidate entities, and computes a dense subgraph that approximates the best joint mention-entity mapping. Experiments show that the new method significantly outper- forms prior methods in terms of accuracy, with robust behavior across a variety of inputs. 1 Introduction 1.1 Motivation Web pages, news articles, blog postings, and other Internet data contain mentions of named entities such as people, places, organizations, etc. Names are often ambiguous: the same name can have many different meanings. For example, given a text like “They per- formed Kashmir, written by Page and Plant. Page played unusual chords on his Gibson.”, how can we tell that “Kashmir” denotes a song by Led Zeppelin and not the Himalaya region (and that Page refers to guitarist Jimmy Page and not to Google founder Larry Page, and that Gibson is a guitar model rather than the actor Mel Gibson)? Establishing these mappings between the mentions and the actual entities is the problem of named-entity disambiguation (NED). If the possible meanings of a name are known up- front - e.g., by using comprehensive gazetteers such as GeoNames (www.geonames.org) or knowledge bases such as DBpedia (Auer07), Freebase (www. freebase.com), or YAGO (Suchanek07), which have harvested Wikipedia redirects and disambigua- tion pages - then the simplest heuristics for name res- olution is to choose the most prominent entity for a given name. This could be the entity with the longest Wikipedia article or the largest number of incoming links in Wikipedia; or the place with the most inhab- itants (for cities) or largest area, etc. Alternatively, one could choose the entity that uses the mention most frequently as a hyperlink anchor text. For the example sentence given above, all these techniques would incorrectly map the mention “Kashmir” to the Himalaya region. We refer to this suite of methods as a popularity-based (mention-entity) prior. Key to improving the above approaches is to con- sider the context of the mention to …

Content truncated to 500 words